How Your Brain Organizes Information

The brain’s ability to generalize across wildly different situations may depend on a flexible “cognitive map” that organizes both physical space and abstract task variables into a shared, reusable framework. That framework matters because it turns one-off experiences—like learning lasagna in a specific kitchen—into transferable knowledge: the same underlying procedure can be applied in a friend’s unfamiliar layout after the brain strips away irrelevant sensory details and plugs the new context into an internal model of how kitchens work.

Behavioral experiments laid the groundwork for this idea long before neuroscientists could observe the underlying neural machinery. In a classic maze study associated with Edward Tolman, rats trained in one maze were later placed in a new maze with multiple radial arms. When a previously rewarded route was blocked, rats often chose paths that pointed toward the goal direction even if those exact routes had never been directly associated with reward—consistent with the presence of an internal spatial representation rather than pure stimulus-reward association. The modern twist is that similar representational principles appear in neural activity across both spatial and non-spatial tasks.

Within the hippocampal formation—especially the hippocampus and entorhinal cortex—research points to a division of labor that supports map-like computation. Place cells in the hippocampus fire when an animal occupies particular locations, but their selectivity depends on context. Grid cells in entorhinal cortex fire in stable, periodic patterns arranged on a hexagonal lattice as an animal moves, providing a kind of coordinate system. Other specialized cells—boundary cells, head direction cells, and landmark- and object-vector-like responses—add information about edges, orientation, and salient features. Crucially, this selectivity is not limited to physical space. Hippocampal neurons can become selective for a particular sound frequency in a one-dimensional “space” of auditory inputs, and entorhinal activity can show grid-like periodicity even when the “environment” is an abstract two-dimensional space defined by bird silhouette parameters.

A unifying computational theme links these diverse domains: represent the world as a graph of connected states, then perform “path integration” on that graph. In physical settings, path integration uses self-motion cues to update position; in abstract settings, the same logic can be applied using rules for how relationships compose. This graph view also clarifies why the same neural machinery could support navigation in rooms, movement in conceptual spaces, and even reasoning over structured categories like family trees.

The brain also appears to infer latent spaces—hidden variables not directly signaled by sensory cues. In a T-maze alternation task, rats must track both physical location and whether the next choice should be left or right based on the previous trial. Neurons dubbed “splitter cells” encode positions in an expanded representation that includes this inferred left/right dimension. In a virtual reality tower accumulation task, hippocampal neurons form place fields over a latent evidence axis defined by the difference in tower counts on each side.

Finally, the system seems to factor knowledge into a structural backbone and a sensory overlay. Entorhinal cortex supplies a stable structural coordinate stream (notably from medial regions with grid-like activity) alongside a sensory stream (from lateral entorhinal areas). The hippocampus then binds these into conjunctive representations, producing context-dependent remapping of place cells when sensory conditions change. The result is a map-like, factorized representation that can generalize efficiently—reusing learned structure while updating sensory context—so the same relational knowledge can guide behavior in new kitchens, new mazes, and new abstract worlds.